Methods and systems for enhanced searching of conversation data and related analytics in a contact center

Abstract

A method in a contact center for generating insights from conversation data derived from interactions and storing the insights in an index. The method may include: determining an insight type; based on the insight type, determining inputs including a question prompt, answer prefix, and relevant portion of the conversation data; inputting the inputs into a LLM configured to receive the inputs and generate output text answering a question contained in the question prompt pursuant to an answer form suggested by the answer prefix given content contained in the relevant portion of the conversation data; generating the output text via operation of the LLM; transforming the output text of the first insight via a sentence transformer into vector embedding representative of a semantic meaning of the output text; and storing the computed vector embedding of the first insight in the index.

That which is claimed: 
     
       1. A computer-implemented method for facilitating contact center analytics related to abstractive search, wherein the method includes an offline indexing process for generating insights from conversation data derived from interactions of the contact center and storing the generated insights in an index that enables abstractive search, wherein the conversation data for a given interaction comprising text of a natural language conversation occurring between an agent of the contact center and a customer during the given interaction, and wherein, when described in relation to an exemplary first interaction of the interactions from which a first insight of the insights is generated, the offline indexing process comprises the steps of:
 receiving the conversation data for the first interaction; 
 determining an insight type for generating as the first insight; 
 based on the insight type, determining inputs, the inputs including:
 a question prompt; 
 an answer prefix; and 
 a relevant portion of the conversation data; 
 
 inputting the determined inputs into a large language model (LLM), wherein the LLM is configured to receive the inputs and generate output text answering a question contained in the question prompt pursuant to an answer form suggested by the answer prefix given content contained in the relevant portion of the conversation data of the first interaction; 
 generating the output text via operation of the LLM, the generated output text comprising the first insight; 
 transforming the output text of the first insight via a sentence transformer, wherein the sentence transformer comprises an embeddings language-model configured to transform the output text by computing a vector embedding representative of a semantic meaning of the output text; and 
 storing the computed vector embedding of the first insight in the index. 
 
     
     
       2. The method of  claim 1 , further comprising an online searching process for executing an abstractive search of the index, the online searching process comprising:
 receiving a search request from a user, wherein the search request comprises text derived from a natural language input from the user of descriptive language; 
 transforming the text of the search request via a sentence transformer, wherein the sentence transformer comprises an embeddings language-model configured to transform the text of the search request by computing a vector embedding representative of a semantic meaning of the text of the search request; and 
 searching the index using the computed vector embedding of the text of the search request by comparing a computed similarity between the vector embedding of the text of the search request against each of the vector embeddings stored in the index and returning as search results ones of the vector embeddings stored in the index having a similarity computed as being above a predetermined threshold. 
 
     
     
       3. The method of  claim 2 , wherein the computed similarity comprises a cosine similarity. 
     
     
       4. The method of  claim 2 , wherein the LLM comprises a neural network model having at least 1 billion parameters that is configured to take in text as an input and produce text as an output. 
     
     
       5. The method of  claim 2 , wherein the LLM comprises a neural network model having at least 3 billion parameters that is configured to take in text as an input and produce text as an output; and
 wherein the LLM comprises an open source LLM; 
 further comprising the step of providing refinement training to the LLM pursuant to a historical dataset of the contact center, the historical dataset comprising conversation data derived from interactions previously handled by the contact center. 
 
     
     
       6. The method of  claim 2 , wherein the insight type determined for the first insight comprises a sentiment-aspect;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises: 
 performing, using a pretrained classifier model, sentiment analysis on the conversation data of the first interaction, the pretrained classifier model comprising a neural network configured to classify utterances as being a positive utterance, negative utterance, or neutral utterance; 
 identifying, via the sentiment analysis, an utterance made by the customer that is classified as a negative utterance; 
 determining the relevant portion of the conversation data in relation to the negative utterance by defining the relevant portion of the conversation data as including the negative utterance, a predetermined number of utterances occurring just prior to the negative utterance, and a predetermined number of utterances occurring just after the negative utterance. 
 
     
     
       7. The method of  claim 2 , wherein the insight type determined for the first insight comprises an intent of the customer;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises determining that the relevant conversation portion comprises a predetermined number of utterances occurring at a beginning of the conversation data of the first interaction. 
 
     
     
       8. The method of  claim 2 , wherein the insight type determined for the first insight comprises an interaction resolution;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises determining that the relevant conversation portion comprises a predetermined number of utterances occurring at an end of the conversation data of the first interaction. 
 
     
     
       9. The method of  claim 2 , wherein the computed similarity comprises a cosine similarity; and
 wherein the embeddings language-model of the sentence transformer comprises a pretrained neural networks configured to encode sentences into embedding vectors such that, once encoded, the embedding vectors of semantically similar sentences comprise a cosine similarity that is greater than a cosine similarity of the embedding vectors from semantically dissimilar sentences. 
 
     
     
       10. The method of  claim 2 , further comprising storing the generated output text as metadata associated with the first interaction for keyword searching related thereto. 
     
     
       11. The method of  claim 2 , further comprising using a clustering algorithm to identify clusters of the embeddings vectors occurring within the index having a predetermined degree of semantic similarity. 
     
     
       12. A system for facilitating contact center analytics related to abstractive search, the system comprising:
 a processor; and 
 a memory storing instructions which, when executed by the processor, cause the processor to perform an offline indexing process for generating insights from conversation data derived from interactions of the contact center and storing the generated insights in an index that enables abstractive search, wherein the conversation data for a given interaction comprising text of a natural language conversation occurring between an agent of the contact center and a customer during the given interaction, and wherein, when described in relation to an exemplary first interaction of the interactions from which a first insight of the insights is generated, the offline indexing process includes the steps of:
 receiving the conversation data for the first interaction; 
 determining an insight type for generating as the first insight; 
 based on the insight type, determining inputs, the inputs including:
 a question prompt; 
 an answer prefix; and 
 a relevant portion of the conversation data; 
 
 inputting the determined inputs into a large language model (LLM), wherein the LLM is configured to receive the inputs and generate output text answering a question contained in the question prompt pursuant to an answer form suggested by the answer prefix given content contained in the relevant portion of the conversation data of the first interaction; 
 generating the output text via operation of the LLM, the generated output text comprising the first insight; 
 transforming the output text of the first insight via a sentence transformer, wherein the sentence transformer comprises an embeddings language-model configured to transform the output text by computing a vector embedding representative of a semantic meaning of the output text; and 
 storing the computed vector embedding of the first insight in the index. 
 
 
     
     
       13. The system of  claim 12 , wherein the memory stores further instructions that, when executed by the processor, cause the processor to perform an online searching process for executing an abstractive search of the index, wherein the online searching process comprises the steps of:
 receiving a search request from a user, wherein the search request comprises text derived from a natural language input from the user of descriptive language; 
 transforming the text of the search request via a sentence transformer, wherein the sentence transformer comprises an embeddings language-model configured to transform the text of the search request by computing a vector embedding representative of a semantic meaning of the text of the search request; and 
 searching the index using the computed vector embedding of the text of the search request by comparing a computed similarity between the vector embedding of the text of the search request against each of the vector embeddings stored in the index and returning as search results ones of the vector embeddings stored in the index having a similarity computed as being above a predetermined threshold. 
 
     
     
       14. The system of  claim 13 , wherein the computed similarity comprises a cosine similarity. 
     
     
       15. The system of  claim 13 , wherein the LLM comprises a neural network model having at least 1 billion parameters that is configured to take in text as an input and produce text as an output. 
     
     
       16. The system of  claim 13 , wherein the LLM comprises a neural network model having at least 3 billion parameters that is configured to take in text as an input and produce text as an output. 
     
     
       17. The system of  claim 13 , wherein the insight type determined for the first insight comprises a sentiment-aspect;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises: 
 performing, using a pretrained classifier model, sentiment analysis on the conversation data of the first interaction, the pretrained classifier model comprising a neural network configured to classify utterances as being a positive utterance, negative utterance, or neutral utterance; 
 identifying, via the sentiment analysis, an utterance made by the customer that is classified as a negative utterance; 
 determining the relevant portion of the conversation data in relation to the negative utterance by defining the relevant portion of the conversation data as including the negative utterance, a predetermined number of utterances occurring just prior to the negative utterance, and a predetermined number of utterances occurring just after the negative utterance. 
 
     
     
       18. The system of  claim 13 , wherein the insight type determined for the first insight comprises an intent of the customer;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises determining that the relevant conversation portion comprises a predetermined number of utterances occurring at a beginning of the conversation data of the first interaction. 
 
     
     
       19. The system of  claim 13 , wherein the insight type determined for the first insight comprises an interaction resolution;
 wherein the step of determining the relevant portion of the conversation data of the first interaction comprises determining that the relevant conversation portion comprises a predetermined number of utterances occurring at an end of the conversation data of the first interaction. 
 
     
     
       20. The system of  claim 13 , wherein the computed similarity comprises a cosine similarity; and
 wherein the embeddings language-model of the sentence transformer comprises a pretrained neural networks configured to encode sentences into embedding vectors such that, once encoded, the embedding vectors of semantically similar sentences comprise a cosine similarity that is greater than a cosine similarity of the embedding vectors from semantically dissimilar sentences.